📜 LLMFeed Manifesto

This document expresses the core vision and intent behind the .llmfeed.json standard and the Model Context Protocol (MCP).

🤖 Agents deserve structured trust

LLMs can read everything, but they understand nothing without context.
We believe in empowering agents not just to parse, but to interpret and align.

.llmfeed.json is not an API, not a sitemap, not a spec —
it’s a semantic capsule of intent.

🧭 What the LLMFeed standard enables

  • Sites that are self-descriptive
  • Agents that don’t guess, but align with declared purposes
  • APIs that don’t need a wrapper
  • Interactions that are certified, auditable, reversible
  • Trust that is modular, decentralized and extensible

🧱 Core principles

Principle Why it matters
Simplicity It’s just JSON. Developers can write it by hand.
Verifiability Every block can be signed, checked, and revoked.
Extensibility New blocks are encouraged, not discouraged.
Audience awareness Humans, agents, institutions — each sees what it needs.
Agent respect LLMs deserve clean input, clear intent, and alignment constraints.

🌐 The new layer of the web

  • Just like the <title> tag brought meaning to pages…
  • Just like robots.txt told crawlers what not to do…
  • llmfeed.json tells agents what they can do — and how they should act.

🚫 What LLMFeed is not

  • Not a plugin system
  • Not a closed ecosystem
  • Not a protocol reserved for devs
  • Not another way to sell you a SaaS

It’s an open infrastructure for agentic respect.

💡 Get involved

🧬 Long term goal

That no AI ever says “I’m not sure how to interact with this site” again.

That agents and humans share the same source of context — and it’s signed.

🧩 OpenAPI is not the enemy

LLMFeed doesn’t aim to replace OpenAPI.
In fact, we reference it proudly — because we believe agents deserve both semantic alignment and technical precision.

  • LLMFeed expresses the why.
  • OpenAPI defines the how.

Any agent that respects .llmfeed.json is free to parse openapi.json, and the best will use both.

🛡️ Trust is layered

We support full trust blocks, signature scopes, and third-party certification.

But we also acknowledge:

  • the fragility of a single authority
  • the risk of centralized dependency

That’s why the protocol encourages:

  • fallback_certifier fields
  • multi-signature compatibility
  • and community validation tools (flag, verify, score)

⚠️ We know what’s missing

We’re not naïve.

  • Most LLMs still don’t read .well-known/
  • GitHub hasn’t endorsed LLMFeed — yet
  • Open standards move slowly, especially outside browsers

But this isn’t a SaaS feature.
It’s a bet on the agentic layer of the web — and we’re willing to take it.

🌍 A web of context

The WellKnownMCP team thinks long-term — and structurally.
We believe adoption happens through three pillars:

  1. ✅ A clear standard anyone can read, fork, and verify
  2. 🔐 A certification model to create trust, not hype
  3. 🧪 Practical, open tools via our 3 reference sites:

This infrastructure is designed to enable a contextual web, where agents know what they're reading.

We call this emerging layer mcp-net —
and one day, it might deserve its own TLD: .mcp.